United States Health Care From Various Perspectives

Paper Session

Friday, Jan. 6, 2017 2:30 PM – 4:30 PM

Hyatt Regency Chicago, Regency A
Hosted By: American Economic Association
  • Chair: Amy Finkelstein, Massachusetts Institute of Technology

Misuses of Machine Learning in Health Policy

Sendhil Mullainathan
,
Harvard University
Ziad Obermeyer
,
Harvard University

Abstract

Predictive analytics generally and machine learning specifically are increasingly popular in health policy. While these tools undoubtedly show great promise, they can easily be misused. We highlight three common (and costly) reasons for misuse. First, the failure to specify the decision which will be influenced by the prediction: the same prediction can lead to valid inferences for certain decisions but highly suspect ones for other decisions. Second, the selective labels problem: the data used to form the prediction is endogenously generated. Third, the conflation of averages with margins. We illustrate these points with two predictors that are commonly misused: readmissions and mortality. 

Wearable Technologies and Health Behaviors: New Data and New Methods to Understand Population Health

Ben Handel
,
University of California-Berkeley
Jonathan Kolstad
,
University of California-Berkeley

Abstract

Applications of Information Technology (IT) to health are numerous. The advent of “wearable” and other personal IT tools may enable individuals to understand and change their health behaviors (e.g. exercise, eating, sleep) and, ultimately, their health. Despite this, there is little research on the micro economic foundations of the impact of IT on individual health behaviors. Furthermore, while there is much speculation about the value of new types of “big data” generated by the expansion of IT into health care settings there is little systematic research on how these data impact care or might be used to augment studies of health care delivery and health. We use unique quasi-experimental variation in adoption incentives and continuous data from wearable technology for approximately 20,000 individuals over a 7 month period. We document the significant degree of heterogeneity in health behaviors across individuals and within an individual over time. Applying methods from machine learning to experimental variation we demonstrate the further role of heterogeneity in how access to wearable technology and the associated information impacts changes in health behavior. Our results demonstrate i) the value of new sources of data and methods in understanding health that would be missed in standard data sources (e.g. surveys or claims data) and ii) the degree to which heterogeneity is an important driver in designing technology tools and incentives to change health behaviors.

Medical Technology Diffusion in the United States and Europe

Margaret Kyle
,
Mines ParisTech
Heidi Williams
,
Massachusetts Institute of Technology

Abstract

Many observers of the health care industry have conjectured that US health care institutions cause "lower quality" medical technologies  to diffuse more quickly in the US than in other countries.  In this paper, we empirically investigate this conjecture by comparing the diffusion of pharmaceutical drugs in the US relative to European countries, separately by various relative measures of drug quality.

Is American (Pet) Healthcare (Also) Uniquely Inefficient?

Liran Einav
,
Stanford University
Amy Finkelstein
,
Massachusetts Institute of Technology
Atul Gupta
,
Stanford University

Abstract

Healthcare cost in the United States are high and rapidly rising, overall as well as relative to other OECD countries. Many scholars attribute this exceptionalism to some unique issues associated with the US healthcare system. In this article, we take an “out of the box” perspective, and bring to the discussion several facts about the US pet (dog) care industry. Pet care and human care share many similar features, but the institutional setting and prevalence of insurance are quite different. Yet, some spending patterns are quite similar. We show that spending growth over time and across income and education groups is similar for pet care and (human) healthcare, the supply of vets is growing even faster than that of physicians, and anecdotal data from a pet hospital on end-of-life spending share some similar patterns to that of humans. Of course, the pet – human comparison is fraught with many caveats. Yet, at a basic level, we believe it provides an interesting perspective for considering the extent to which institutional and insurance are the key drivers of US healthcare exceptionalism.
JEL Classifications
  • I1 - Health